In IBM PowerVM on systems built with the IBM POWER architecture, shared processor pool (SPP) scheduling is how the hypervisor (PHYP) distributes CPU time among micro-partitioned LPARs that share a defined set of physical cores.

It’s a fair-share, priority-aware, time-sliced scheduling system with dynamic borrowing.

🔹 1. What a Shared Processor Pool Is

A Shared Processor Pool (SPP) is:

• A group of physical CPU cores
• Shared by multiple LPARs

Each LPAR in the pool has:

• Entitled capacity (guaranteed CPU)
• Optional ability to use more (uncapped mode)

🔹 2. Core Scheduling Concepts

🧠 a) Entitled Capacity (EC)

• Minimum guaranteed CPU for an LPAR
• Example:
  • 0.5 cores guaranteed

⚖️ b) Virtual Processors (VPs)

• Logical CPUs presented to the LPAR
• Determines how much parallel work it can run

🚦 c) Capped vs Uncapped

• Capped → cannot exceed entitlement
• Uncapped → can use extra CPU if available

🧩 d) Weight (Priority)

• Used in uncapped mode
• Determines how extra CPU is shared

🔹 3. How Scheduling Works (Step-by-Step)

🟢 Step 1: Guarantee Phase

• PHYP first ensures all LPARs get their entitled capacity

👉 This is strict and guaranteed

🟡 Step 2: Surplus Distribution

• If unused CPU exists:
  • Distributed among uncapped LPARs

👉 Based on:

• Weight
• Demand

🔴 Step 3: Time-Slice Scheduling

• CPU time divided into small time slices
• PHYP rotates execution across LPARs

👉 Very fast switching (microseconds level)

🔹 4. Example Scenario

Pool: 4 physical cores

Allocation:

1. Guaranteed:
   • A = 1 core
   • B = 1 core
   • C = 0.5 core

👉 Total = 2.5 cores used

2. Remaining = 1.5 cores

• Shared between A and B based on weight:

👉 A gets ~2/3
👉 B gets ~1/3

🔹 5. Hardware Support

On processors like the IBM POWER10 processor:

• SMT threads allow fine-grained sharing
• Fast context switching reduces overhead
• Hardware counters help scheduler decisions

🔹 6. Performance Characteristics

✅ High Utilization

• Idle CPU is never wasted

⚡ Fast Response

• Microsecond-level scheduling

⚖️ Fairness

• Guaranteed minimum + weighted sharing

🔄 Dynamic Adaptation

• Adjusts in real time based on workload demand

🔹 7. Potential Trade-offs

❌ Latency variability

• Shared environment → contention possible

❌ Overcommit risk

• Too many LPARs → CPU contention

❌ Tuning complexity

• Requires correct:
  • Entitlement
  • Weight
  • VP sizing

🔹 8. Best Practices

✔ Match virtual processors to workload parallelism
✔ Avoid excessive overcommit
✔ Use uncapped mode for bursty workloads
✔ Monitor CPU wait time

🔹 9. Key Insight

Shared processor pool scheduling ensures guaranteed performance first, then maximizes utilization through intelligent sharing.

🔑 Summary

🧠 Bottom Line

In IBM PowerVM:

• Shared processor pools use fair-share scheduling with guarantees
• Combine:
  • Deterministic performance
  • High efficiency

👉 This allows POWER systems to achieve very high consolidation ratios without sacrificing critical workload performance.